Tcl Source Code

# Create namespaces
namespace eval a { namespace eval b {} }

# Good result with non-glob pattern, and global namespace
namespace children :: a  ;# => ::a

# Good result with single-char glob pattern, and other namespace
namespace children a {[b]}  ;# => ::a::b

# Good result because of glob char in pattern, and other namespace
namespace children a b*  ;# => ::a::b

# Bad result with non-glob pattern, and other namespace
namespace children a b  ;# => {}


In generic/tclNamespace.c, function NamespaceChildrenCmd (lines 2988-3108), Tcl_StringMatch will be called if a glob char is present in the pattern, otherwise the matching will be performed with a combination of strncmp() and Tcl_FindHashEntry:
line 3067:  if ((pattern != NULL) && TclMatchIsTrivial(pattern)) {

The hash table lookup:
line 3075 or 3078:  Tcl_FindHashEntry(&nsPtr->childTable, pattern+length) != NULL

These two chars might be added but they won't be taken into account when matching: lines 3053-3055:
if (nsPtr != globalNsPtr) {
    TclDStringAppendLiteral(&buffer, "::");
}


With a quick hack patch:
-Tcl_FindHashEntry(&nsPtr->childTable, pattern+length) != NULL
+Tcl_FindHashEntry(&nsPtr->childTable, 2+pattern+length) != NULL

The child namespace can now be found:
namespace children a b  ;# => ::a::b

However namespace ::a can no longer be found in the global namespace!
namespace children :: a  ;# => {}


The two namespace separator ("::") chars are not being accounted for properly.

Backport landed on the 8.6 branch last December, so I guess
the decision got made.  Closing.

Great ! I would love some explanatory text around the fix source. Maybe, I care tomorrow !

And it is 9.0.1, not 9.0.2, right ?

Take care, Harald

The CI build shows no regressions. So, Yes I think this should be included into Tcl 9.0.2 (to be discussed within the TTT).

Was this problem introduced in Tcl 8.7 or 9.0, or was it always there? Should it be backported to 8.7 and/or 8.6? Therefore, leaving this ticket open. No hurry to decide that.

Proposed fix [ee821ec00f|here]

The patch is just a suggestion.
The problem is an offset of 2 is not added when it should be.
I'm not familiar enough with the code to pick the best place to do this.

Correction:
-Tcl_FindHashEntry(&nsPtr->childTable, pattern+length) != NULL
+Tcl_FindHashEntry(&nsPtr->childTable, (nsPtr != globalNsPtr ? 2 : 0)+pattern+length) != NULL

This patch handles global and non-global. q&d poc

-Tcl_FindHashEntry(&nsPtr->childTable, 2+pattern+length) != NULL
+Tcl_FindHashEntry(&nsPtr->childTable, (nsPtr != globalNsPtr ? 2 : 0)+pattern+length) != NULL